Track of Implemented Features

This issue serves to track implemented features in the latest release of Ianvs.

Ianvs includes a wide range of features showcasing its benchmarking capablilities. We summarize the list here to track features with implementation.

To anyone is interested in contributing to these features: Note that for these important features, a proposal pull request is needed to show the design of before implementation. Then a formal presentation shall be given in the KubeEdge SIG AI to launch a review of architecture design before getting to any implementation. See a proposal example in https://github.com/kubeedge/ianvs/blob/main/docs/proposals/scenarios/GovDoc2Poster/GovDoc2Poster.md

Call for Contributions: Not Yet Tackled

• Parallel processing Not supported parallel processing of multiple use cases yet
• Name consistence Inconsistent with interface name 'initial_model_url' and environment variable 'base_model_url'
• CICD for doc spell check Fix various spelling errors and typos in the documentation and code comments
• Privacy-Preserving LLM Phase 3: Multimodal Extension Feature Request: Multimodal Extension for PIPL-Compliant Privacy-Preserving LLM
• LLM Speculative Decoding Phase 3 [LFX 2026 term1] Cloud-Edge Simulation Benchmark for LLM Speculative Decoding in KubeEdge-Ianvs

Call for Improvement: Have Merged Attempts & Could Enhanced

Simulator for complicated edge-cloud system

• Simulator for Edge-cloud Collaborative AI: a Lifelong Learning example https://github.com/kubeedge/ianvs/issues/348

Single task learning

• Edge LLM Large Language Model Edge Benchmark Suite: Implementation on KubeEdge-Ianvs
• Coding agent Smart Coding benchmark suite: built on KubeEdge-lanvs
• Personalized agent Personalized LLM Agent based on KubeEdge-Ianvs cloud-edge collaborative lifelong learning
• Multimodal LM Multimodal Large Model Joint Learning Algorithm: Reproduction Based on KubeEdge-Ianvs
• Embodied intelligence Embodied Intelligence Benchmarking Framework for Industrial Manufacturing with KubeEdge-Ianvs
• Data Transfer Learning for VLA Data Mixture Optimization Algorithm for VLA Fine-Tuning Based on KubeEdge-Ianvs
• Privacy-Preserving LLM Phase 2: Scheme Optimizing Privacy-Performance Equilibrium in Cloud-Edge LLM Systems
• Domain-specific LM Phase 3: Gov Agent Agent Benchmark for Government Scenarios Based on KubeEdge-Ianvs

Joint inference

• LLM joint inference for semantic segmenation Real-Time IoT Perception Systems Based on Edge-Cloud Collaboration with Large Foundation Models
• Multiedge inference Heterogeneous Multiedge Inference for High Mobility Scenarios
• Multi-LLM routing Cloud-edge multi-LLM routing paradigm based on KubeEdge-Ianvs: Confronted with Embodied Intelligence

Federated learning

• Federated incremental learning Federated Incremental Learning for Label Scarcity Scenarios based on KubeEdge-Ianvs

Lifelong learning

• Class incremental semantic segmentaion Implementation of a Class Incremental Learning Algorithm Evaluation System based on Ianvs

Completed & Closed

Single task learning

• Scene Generation Physically Consistent and Interactable Indoor Simulation Scene Generation: Implementation Based on KubeEdge-Ianvs
• Domain-specific LM Phase 1: Gov LLM Domain-Specific Large Model Benchmarking Based on KubeEdge-Ianvs
• Domain-specific LM Phase 2: Gov RAG Domain-specific large model benchmarks: the edge perspective

Joint inference

• LLM Speculative Decoding Phase 1 Cloud-edge collaborative inference for LLM based on KubeEdge-Ianvs
• LLM Speculative Decoding Phase 2 Cloud-edge collaborative speculative decoding for LLM based on KubeEdge-Ianvs
• Privacy-Preserving LLM Phase 1: Metrics Privacy-Preserving Prompt Processing Framework for Cloud-Edge Collaborative LLMs on KubeEdge-Ianvs

Federated learning

• Federated LLM Large Language Model Federated Fine-tuning Algorithm Based on KubeEdge-Ianvs

Lifelong learning

• Edge-cloud collaborative lifelong learning for cloud robotics Cloud-Robotic AI Benchmarking for Edge-cloud Collaborative Lifelong Learning
• Automatic Image Annotation for lifelong learning Automatic Image Annotation Algorithm based on Sedna Lifelong Learning to reproduce open world object Segmentation